P1928R12

P1928/changelog.tex

@@ -85,7 +85,7 @@ \section{Changelog}
 \item Use \code{indirectly_writable} instead of \code{output_iterator}.
 \item Replace most \code{size_t} and \code{int} uses by \code{\textit{simd-size-type}} signed integer type.
 \item Remove everything in \code{simd_abi} and the namespace itself.
-\item Reword section on ABI tags using exposition-only ABI tag aliases.
+\item Reword section on ABI tags using exposition only ABI tag aliases.
 \item Guarantee generator ctor calls callable exactly once per index.
 \item Remove \code{int}/\code{unsigned int} exception from conversion rules of broadcast ctor.
 \item Rename \code{loadstore_flags} to \code{simd_flags}.
@@ -97,7 +97,7 @@ \section{Changelog}
 \item Rename \code{simd<T, Abi>} to \code{basic_simd<Bytes, Abi>} and replace all occurrences accordingly.
 \item Remove \code{long double} from the set of vectorizable types.
 \item Remove \code{is_abi_tag}, \code{is_simd}, and \code{is_simd_mask} traits.
-\item Make \code{simd_size} exposition-only.
+\item Make \code{simd_size} exposition only.
 \end{revision}
 
 \begin{revision}
@@ -119,7 +119,7 @@ \section{Changelog}
 \begin{revision}
 \item Include \code{std::optional} return value from \code{reduce_min_index} and \code{reduce_max_index} in the exploration.
 \item Fix \LaTeX{} markup errors.
-\item Remove repetitive mention of “exposition-only” before \deducet.
+\item Remove repetitive mention of “exposition only” before \deducet.
 \item Replace “TU” with “translation unit”.
 \item Reorder first paragraphs in the wording, especially reducing the note on compiling down to SIMD instructions.
 \item Replace cv-unqualified arithmetic types with a more precise list of types.
@@ -139,7 +139,7 @@ \section{Changelog}
 \item Add instructions to add \code{<simd>} to the table of headers in [headers].
 \item Add instructions to add a new subclause to the table in [numerics.general].
 \item Add instructions to add \code{<simd>} [diff.23.library].
-\item Add \simdsizev to the wording and replace \code{simd_size_v} to actually implement “Make \code{simd_size} exposition-only.”
+\item Add \simdsizev to the wording and replace \code{simd_size_v} to actually implement “Make \code{simd_size} exposition only.”
 \item Restored precondition (and removed \code{noexcept}) on
   \code{reduce_min_index} and \code{reduce_max_index} as directed by LEWG.
 \end{revision}
@@ -173,7 +173,7 @@ \section{Changelog}
 \item Fix constraints on shift operators with \simdsizetype{} on the right operand.
 \item Remove wording removed by P3275 (non-const \code{operator[]}).
 \item Make intrinsics conversion recommended practice.
-\item Make \code{simd_flags} template arguments exposition-only.
+\item Make \code{simd_flags} template arguments exposition only.
 \item Make \code{simd_alignment} \emph{not} implementation-defined.
 \item Reword “supported” to “enabled or disabled”.
 \item Apply improved wording from \ref{sec:simd.overview} to \ref{sec:simd.mask.overview}.
@@ -182,6 +182,52 @@ \section{Changelog}
 \item Use \code{to_address} on contiguous iterators (\ref{sec:simd.ctor} and \ref{sec:simd.copy}).
   This is more explicit about allowing memcpy on the complete range rather than
   having to iterate the range per element.
+\end{revision}
+
+\begin{revision}
+\item Fix default size of \code{simd} and \code{simd_mask} aliases
+  (\ref{sec:simd.syn}, necessary for
+  \std\code{destructible<\MayBreak{}\std{}simd<\MayBreak\std{}string>>} to be well-formed).
+\item Extend value-preserving to encompass conversions from all arithmetic
+  types. Use this new freedom in \ref{sec:simd.ctor} to fully constrain the
+  generator constructor and to plug a specification hole in the broadcast
+  constructor.
+\item Fix broadcast constructor wording by constraining \constexprwrapperlike
+  arguments to arithmetic types.
+\item Reword the note in \ref{sec:simd.overview} and
+  \ref{sec:simd.mask.overview} to talk about the “architecture of the execution
+  environment” as used in the note on \code{int} in [basic.fundamental].
+\item Say \code{binary_op} instead of \code{BinaryOperation} in “\ldots does
+  not modify \code{x}”. (\ref{sec:simd.reductions})
+\item Improve presentation of inputs given to
+  \code{\placeholdernc{GENERALIZED_SUM}} in \ref{sec:simd.reductions}.
+\item Fix use of the simd broadcast constructor in precondition of
+  \code{reduce}.
+\item Drop the notes that were trying to explain the intent behind
+  enabling/disabling simd(mask) specializations.
+\item Reorder \code{binary_op} and \code{identity_element} on masked
+  \code{reduce}. Provide \code{std::plus<>} as default \code{BinaryOperation}
+  and specify a default argument for \code{identity_element}
+  (\ref{sec:simd.reductions}).
+\item Use “Equivalent to” wording for \code{simd_select} and carve out an
+  exception for ADL for \simdselect (\ref{sec:simd.alg}).
+\item Add a constraint for a provided identity element argument for masked
+  reduce if the binary operation isn't one of the five known operations.
+\item Add exposition only concept \reductionoperation with additional semantic
+  constraints and use it in the \code{reduce} constraints
+  (\ref{sec:simd.reductions}).
+\item Simplify precondition on identity element, using only \code{simd<T, 1>}
+  instead of all possible ABI tags. The latter follows from the former through
+  mandating an \emph{element-wise binary operation}.
+\item Add constraints for \code{simd_split} and \code{simd_cat}, requiring
+  enabled \code{basic_simd}/\code{basic_simd_mask} types
+  (\ref{sec:simd.creation}).
+\item Specify \code{minmax} using “Equivalent to:” wording.
+  (\ref{sec:simd.alg})
+\item Discuss whether we could/should make \code{simd_flags} exposition only.
+  (\sect{sec:simd_flags-expos})
+\item Move all exposition only types, variables, and concepts into their own
+  section before the synopsis. (\ref{sec:simd.expos})
 
   %\todo Reorder \code{simd} and \code{simd_mask} specification in the wording (mask first).
 \end{revision}

P1928/main.tex

@@ -1,6 +1,6 @@
 \newcommand\wgTitle{std::simd --- merge data-parallel types from the Parallelism TS 2}
 \newcommand\wgName{Matthias Kretz <[email protected]>}
-\newcommand\wgDocumentNumber{P1928R11}
+\newcommand\wgDocumentNumber{P1928R12}
 \newcommand\wgGroup{LWG}
 \newcommand\wgTarget{\CC{}26}
 \newcommand\wgAcknowledgements{Thanks to Daniel Towner, Ruslan Arutyunyan, Jonathan Müller, Jeff Garland, and Nicolas Morales for discussions and/or pull requests on this/previous paper(s).}
@@ -39,6 +39,9 @@
 \newcommand\convertflag{\UNSP{convert-flag}}
 \newcommand\alignedflag{\UNSP{aligned-flag}}
 \newcommand\overalignedflag{\UNSP{overaligned-flag}}
+\newcommand\reductionoperation{\UNSP{reduction-binary-operation}}
+\newcommand\simdfloatingpoint{\UNSP{simd-floating-point}}
+\newcommand\multisimdfloatingpoint{\UNSP{multi-arg-simd-floating-point}}
 
 \renewcommand{\lst}[1]{Listing~\ref{#1}}
 \renewcommand{\sect}[1]{Section~\ref{#1}}
@@ -297,7 +300,7 @@ \subsection{Make use of \code{int} and \code{size_t} consistent}\label{sec:simds
 
 Alignments and values identifying a \code{sizeof} still use \code{size_t}.
 
-The type \simdsizetype{} is an exposition-only alias for a signed integer type.
+The type \simdsizetype{} is an exposition only alias for a signed integer type.
 I.e. the implementation is free to choose any signed integer type.
 
 The rationale given in the LEWG discussion was a desire to avoid type
@@ -493,18 +496,32 @@ \subsection{No freestanding SIMD}\label{sec:freestanding}
 Note that freestanding is just the baseline requirement and embedded targets
 are still free to add \code{simd} support.
 
-\section{Outlook}
-\subsection{\code{element_reference} is overspecified}
-\code{element_reference} is spelled out in a lot of detail.
-It may be better to define its requirements in a list of requirements or a table instead.
+\section{Design clarifications while in LWG review}
+\subsection{Should the \code{simd_flags} template be exposition only?}\label{sec:simd_flags-expos}
 
-This change is not reflected in the wording, pending encouragement from WG21 (mostly LWG).
+The \code{simd_flags} class template could be exposition only in the standard
+without breaking any of the direct use cases of the \simd interface.
+However, if a user needs to write a function that passes \code{simd_flags} into
+a function, how should the function parameter be constrained?
+Do we want everyone to write an unconstrained:
+\medskip\begin{lstlisting}[style=Vc]
+void f(auto flags) { v.copy_to(data, flags); }
+\end{lstlisting}
+or do we want to enable:
+\medskip\begin{lstlisting}[style=Vc]
+template <typename... Flags>
+void f(std::simd_flags<Flags...> flags) { v.copy_to(data, flags); }
+\end{lstlisting}
 
-As an alternative \cite{P3275R0} discusses removal of non-const subscripts
-altogether.
-This would imply removal of \code{element_reference}, simplifying the wording
-by a good chunk.
+If we make the \code{simd_flags} class template exposition only we should
+consider adding a \code{concept} or at least a type trait in its place:
+\medskip\begin{lstlisting}[style=Vc]
+void f(std::simd_flags auto flags) { v.copy_to(data, flags); }
+\end{lstlisting}
+I don't think this is a clear improvement.
+In any case, a change like this would need to go via LEWG.
 
+\section{Outlook}
 \subsection{Clean up math function overloads}
 The wording that produces \simd overloads misses a few cases and leaves room for ambiguity.
 There is also no explicit mention of integral overloads that are supported in \code{<cmath>} (e.g. \std\code{cos(1)} calling \std\code{cos(double)}).
@@ -592,7 +609,7 @@ \subsection{Formatting support}\label{sec:formatting}
 %
 %\subsection{\code{simd_select} overload set}\label{sec:simdselectwording}
 %
-%\wglink{P1928R6} presented the following overload set of the exposition-only
+%\wglink{P1928R6} presented the following overload set of the exposition only
 %hidden friend \code{\simdselect}:
 %\begin{codeblock}
 %template<class T, class Abi> class basic_simd {